AVA,Dx

AVA,Dx analyzes exonic variants from whole exome or whole genome sequencing to predict Crohn's disease (CD) status and identify CD-associated genes and pathogenesis pathways.

Key Features:

• Machine Learning Integration: Employs a machine learning framework that leverages person-specific coding variations to build predictive models and identify known and novel CD-associated genes.
• Exonic Variant Focus: Analyzes exonic variants derived from whole exome or whole genome sequencing as the primary input for model construction.
• Batch Effect Adjustment: Incorporates batch effect adjustments to improve prediction accuracy across separately sequenced panels.
• Performance Metrics: At a strict cutoff, identified 16% of CD patients with 99% precision, and at the default cutoff recognized 58% of patients with 82% precision across over 3,000 individuals from separately sequenced panels.
• Known Gene Recovery: Recovers established CD genes such as NOD2 while suggesting additional potential CD-associated genes.
• Pathogenesis Insight: Facilitates identification of biological pathways and mechanisms associated with Crohn's disease pathogenesis.
• Clinical Assessment Utility: Provides predictive outputs that can be used to assess CD risk and support clinical evaluation and patient stratification.

Scientific Applications:

• Disease Prediction: Predicts individual predisposition to Crohn's disease using exonic variant-derived predictive scores.
• Genetic Research: Identifies candidate CD-associated genes for follow-up genetic and functional studies.
• Pathogenesis Exploration: Supports exploration of pathways and mechanisms implicated in CD etiology and progression.
• Clinical Assessment and Stratification: Supplies predictive information that can inform clinical risk assessment and potential early-intervention strategies.

Methodology:

Constructs predictive models from exonic variant data from whole exome or whole genome sequencing using a machine learning framework that leverages person-specific coding variations; models were trained on panels comprising 111 individuals and include batch effect adjustments.